Model save

README.md

@@ -1,13 +1,11 @@
 ---
+license: apache-2.0
 library_name: peft
 tags:
-- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
 base_model: norallm/normistral-7b-warm
-datasets:
-- hugodk-sch/aftonposten_title_prefs
 model-index:
 - name: ap-normistral-7b-align-scan
   results: []
@@ -18,17 +16,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # ap-normistral-7b-align-scan
 
-This model is a fine-tuned version of [data/ap-normistral-7b-sft-qlora](https://huggingface.co/data/ap-normistral-7b-sft-qlora) on the hugodk-sch/aftonposten_title_prefs dataset.
+This model is a fine-tuned version of [norallm/normistral-7b-warm](https://huggingface.co/norallm/normistral-7b-warm) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.9722
-- Rewards/chosen: -0.0621
-- Rewards/rejected: -0.1987
-- Rewards/accuracies: 0.5453
-- Rewards/margins: 0.1366
-- Logps/rejected: -36.2149
-- Logps/chosen: -32.5208
-- Logits/rejected: 98.875
-- Logits/chosen: 98.9020
+- Loss: 0.9383
+- Rewards/chosen: -0.0593
+- Rewards/rejected: -0.1492
+- Rewards/accuracies: 0.5457
+- Rewards/margins: 0.0899
+- Logps/rejected: -36.2648
+- Logps/chosen: -32.5617
+- Logits/rejected: 98.6998
+- Logits/chosen: 98.7178
 
 ## Model description
 
@@ -63,9 +61,9 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.9567        | 0.26  | 100  | 1.1574          | -0.0132        | 0.0638           | 0.4630             | -0.0769         | -35.8868       | -32.4596     | 98.7233         | 98.7349       |
-| 0.8098        | 0.52  | 200  | 1.0545          | -0.0943        | -0.1571          | 0.5278             | 0.0627          | -36.1629       | -32.5611     | 98.8438         | 98.8682       |
-| 0.6965        | 0.78  | 300  | 0.9869          | -0.0473        | -0.1807          | 0.5831             | 0.1334          | -36.1923       | -32.5023     | 98.8800         | 98.9053       |
+| 0.9172        | 0.26  | 100  | 0.9737          | -0.0075        | -0.0499          | 0.5104             | 0.0424          | -36.0662       | -32.4581     | 98.7856         | 98.7961       |
+| 0.6396        | 0.52  | 200  | 0.9232          | -0.0626        | -0.1726          | 0.5835             | 0.1100          | -36.3118       | -32.5684     | 98.6629         | 98.6844       |
+| 0.6566        | 0.78  | 300  | 0.9383          | -0.0593        | -0.1492          | 0.5457             | 0.0899          | -36.2648       | -32.5617     | 98.6998         | 98.7178       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9279e132994a286000a702418d244bdaff75918026ab34bc32dbeeb431676448
+oid sha256:bc657148de70669dc72c62368af5511f8c847fbd7888765ecc740c6e45452c92
 size 671150064

all_results.json

@@ -1,20 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": 98.90196990966797,
-    "eval_logits/rejected": 98.875,
-    "eval_logps/chosen": -32.520835876464844,
-    "eval_logps/rejected": -36.21490478515625,
-    "eval_loss": 0.9721795320510864,
-    "eval_rewards/accuracies": 0.545265793800354,
-    "eval_rewards/chosen": -0.06213000789284706,
-    "eval_rewards/margins": 0.13657446205615997,
-    "eval_rewards/rejected": -0.19870448112487793,
-    "eval_runtime": 103.7342,
-    "eval_samples": 343,
-    "eval_samples_per_second": 3.307,
-    "eval_steps_per_second": 0.415,
-    "train_loss": 0.7616742146479619,
-    "train_runtime": 2551.1965,
+    "train_loss": 0.735349414874981,
+    "train_runtime": 2550.735,
     "train_samples": 3079,
     "train_samples_per_second": 1.207,
     "train_steps_per_second": 0.151

train_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.7616742146479619,
-    "train_runtime": 2551.1965,
+    "train_loss": 0.735349414874981,
+    "train_runtime": 2550.735,
     "train_samples": 3079,
     "train_samples_per_second": 1.207,
     "train_steps_per_second": 0.151

trainer_state.json

@@ -10,7 +10,7 @@
   "log_history": [
     {
       "epoch": 0.0,
-      "grad_norm": 104.5,
+      "grad_norm": 65.5,
       "learning_rate": 1.282051282051282e-07,
       "logits/chosen": 88.18099975585938,
       "logits/rejected": 88.25153350830078,
@@ -25,628 +25,628 @@
     },
     {
       "epoch": 0.03,
-      "grad_norm": 78.0,
+      "grad_norm": 54.5,
       "learning_rate": 1.282051282051282e-06,
-      "logits/chosen": 81.08700561523438,
-      "logits/rejected": 80.79035186767578,
-      "logps/chosen": -34.29237365722656,
-      "logps/rejected": -33.04549026489258,
-      "loss": 0.9866,
-      "rewards/accuracies": 0.3888888955116272,
-      "rewards/chosen": -0.040518708527088165,
-      "rewards/margins": 0.029149238020181656,
-      "rewards/rejected": -0.06966794282197952,
+      "logits/chosen": 81.0613784790039,
+      "logits/rejected": 80.76800537109375,
+      "logps/chosen": -34.20709228515625,
+      "logps/rejected": -33.1387939453125,
+      "loss": 0.8925,
+      "rewards/accuracies": 0.5416666865348816,
+      "rewards/chosen": 0.017316989600658417,
+      "rewards/margins": 0.10751060396432877,
+      "rewards/rejected": -0.09019361436367035,
       "step": 10
     },
     {
       "epoch": 0.05,
-      "grad_norm": 85.5,
+      "grad_norm": 56.0,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": 80.68280029296875,
-      "logits/rejected": 80.57057189941406,
-      "logps/chosen": -33.53000259399414,
-      "logps/rejected": -30.8519287109375,
-      "loss": 0.8984,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.12792269885540009,
-      "rewards/margins": 0.18747423589229584,
-      "rewards/rejected": -0.05955154448747635,
+      "logits/chosen": 80.64421081542969,
+      "logits/rejected": 80.53112030029297,
+      "logps/chosen": -33.46975326538086,
+      "logps/rejected": -30.782833099365234,
+      "loss": 0.8999,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.11007847636938095,
+      "rewards/margins": 0.11275172233581543,
+      "rewards/rejected": -0.002673250390216708,
       "step": 20
     },
     {
       "epoch": 0.08,
-      "grad_norm": 73.0,
+      "grad_norm": 60.75,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": 82.52570343017578,
-      "logits/rejected": 82.552734375,
-      "logps/chosen": -33.96383285522461,
-      "logps/rejected": -31.202930450439453,
-      "loss": 1.1955,
-      "rewards/accuracies": 0.4000000059604645,
-      "rewards/chosen": 0.03918968886137009,
-      "rewards/margins": -0.1375342458486557,
-      "rewards/rejected": 0.1767239272594452,
+      "logits/chosen": 82.45377349853516,
+      "logits/rejected": 82.48570251464844,
+      "logps/chosen": -33.79710006713867,
+      "logps/rejected": -31.156320571899414,
+      "loss": 1.0372,
+      "rewards/accuracies": 0.42500001192092896,
+      "rewards/chosen": 0.10785794258117676,
+      "rewards/margins": -0.025897562503814697,
+      "rewards/rejected": 0.13375550508499146,
       "step": 30
     },
     {
       "epoch": 0.1,
-      "grad_norm": 74.5,
+      "grad_norm": 47.0,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": 80.97098541259766,
-      "logits/rejected": 80.9694595336914,
-      "logps/chosen": -32.8037109375,
-      "logps/rejected": -33.232460021972656,
-      "loss": 0.9309,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.2962803840637207,
-      "rewards/margins": 0.1874258816242218,
-      "rewards/rejected": 0.10885453224182129,
+      "logits/chosen": 80.95109558105469,
+      "logits/rejected": 80.94834899902344,
+      "logps/chosen": -32.803199768066406,
+      "logps/rejected": -33.15715026855469,
+      "loss": 0.9441,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.1854299008846283,
+      "rewards/margins": 0.07974345237016678,
+      "rewards/rejected": 0.10568644851446152,
       "step": 40
     },
     {
       "epoch": 0.13,
-      "grad_norm": 70.0,
+      "grad_norm": 43.25,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": 78.56432342529297,
-      "logits/rejected": 78.58203125,
-      "logps/chosen": -30.68828773498535,
-      "logps/rejected": -30.789108276367188,
-      "loss": 0.9855,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.4135704040527344,
-      "rewards/margins": 0.18622872233390808,
-      "rewards/rejected": 0.2273416966199875,
+      "logits/chosen": 78.6640625,
+      "logits/rejected": 78.6778335571289,
+      "logps/chosen": -30.60561180114746,
+      "logps/rejected": -30.74114418029785,
+      "loss": 0.9478,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.29981669783592224,
+      "rewards/margins": 0.1337459534406662,
+      "rewards/rejected": 0.16607072949409485,
       "step": 50
     },
     {
       "epoch": 0.16,
-      "grad_norm": 84.0,
+      "grad_norm": 50.5,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": 83.11712646484375,
-      "logits/rejected": 83.17086791992188,
-      "logps/chosen": -30.859710693359375,
-      "logps/rejected": -29.577678680419922,
-      "loss": 0.9236,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.25236082077026367,
-      "rewards/margins": 0.24208447337150574,
-      "rewards/rejected": 0.010276327840983868,
+      "logits/chosen": 83.31804656982422,
+      "logits/rejected": 83.37259674072266,
+      "logps/chosen": -30.898645401000977,
+      "logps/rejected": -29.465993881225586,
+      "loss": 0.9771,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.13825830817222595,
+      "rewards/margins": 0.07599552720785141,
+      "rewards/rejected": 0.06226278096437454,
       "step": 60
     },
     {
       "epoch": 0.18,
-      "grad_norm": 107.0,
+      "grad_norm": 62.5,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": 83.80155944824219,
-      "logits/rejected": 83.8315658569336,
-      "logps/chosen": -30.606204986572266,
-      "logps/rejected": -32.91896057128906,
-      "loss": 1.141,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.01825830340385437,
-      "rewards/margins": -0.0767388790845871,
-      "rewards/rejected": 0.09499720484018326,
+      "logits/chosen": 83.98554229736328,
+      "logits/rejected": 84.01966094970703,
+      "logps/chosen": -30.47078514099121,
+      "logps/rejected": -33.03969192504883,
+      "loss": 0.9425,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.0791207104921341,
+      "rewards/margins": 0.08011289685964584,
+      "rewards/rejected": -0.0009921938180923462,
       "step": 70
     },
     {
       "epoch": 0.21,
-      "grad_norm": 84.5,
+      "grad_norm": 55.5,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": 81.45521545410156,
-      "logits/rejected": 81.44440460205078,
-      "logps/chosen": -31.48111343383789,
-      "logps/rejected": -31.072092056274414,
-      "loss": 0.8995,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.01906520128250122,
-      "rewards/margins": 0.2706204056739807,
-      "rewards/rejected": -0.2515551745891571,
+      "logits/chosen": 81.63408660888672,
+      "logits/rejected": 81.62284088134766,
+      "logps/chosen": -31.412799835205078,
+      "logps/rejected": -30.981252670288086,
+      "loss": 0.8729,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.046071670949459076,
+      "rewards/margins": 0.1578732579946518,
+      "rewards/rejected": -0.11180160194635391,
       "step": 80
     },
     {
       "epoch": 0.23,
-      "grad_norm": 102.0,
+      "grad_norm": 67.0,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": 78.27713775634766,
-      "logits/rejected": 78.24540710449219,
-      "logps/chosen": -32.49984359741211,
-      "logps/rejected": -31.142908096313477,
-      "loss": 0.9467,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.11066800355911255,
-      "rewards/margins": 0.2638159692287445,
-      "rewards/rejected": -0.15314793586730957,
+      "logits/chosen": 78.38566589355469,
+      "logits/rejected": 78.3572998046875,
+      "logps/chosen": -32.54473114013672,
+      "logps/rejected": -31.261743545532227,
+      "loss": 0.8672,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.04672273248434067,
+      "rewards/margins": 0.2018592804670334,
+      "rewards/rejected": -0.15513655543327332,
       "step": 90
     },
     {
       "epoch": 0.26,
-      "grad_norm": 73.5,
+      "grad_norm": 59.25,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": 83.56034851074219,
-      "logits/rejected": 83.58055877685547,
-      "logps/chosen": -34.057029724121094,
-      "logps/rejected": -31.78998374938965,
-      "loss": 0.9567,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.19836829602718353,
-      "rewards/margins": 0.22992074489593506,
-      "rewards/rejected": -0.031552452594041824,
+      "logits/chosen": 83.54986572265625,
+      "logits/rejected": 83.57047271728516,
+      "logps/chosen": -34.15292739868164,
+      "logps/rejected": -31.826343536376953,
+      "loss": 0.9172,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.07603093981742859,
+      "rewards/margins": 0.11392843723297119,
+      "rewards/rejected": -0.0378975048661232,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": 98.73486328125,
-      "eval_logits/rejected": 98.72332763671875,
-      "eval_logps/chosen": -32.45964431762695,
-      "eval_logps/rejected": -35.88682556152344,
-      "eval_loss": 1.1574316024780273,
-      "eval_rewards/accuracies": 0.4630398452281952,
-      "eval_rewards/chosen": -0.013175476342439651,
-      "eval_rewards/margins": -0.07693858444690704,
-      "eval_rewards/rejected": 0.0637631043791771,
-      "eval_runtime": 104.0187,
-      "eval_samples_per_second": 3.297,
+      "eval_logits/chosen": 98.79605865478516,
+      "eval_logits/rejected": 98.78559875488281,
+      "eval_logps/chosen": -32.45810317993164,
+      "eval_logps/rejected": -36.06624221801758,
+      "eval_loss": 0.9737041592597961,
+      "eval_rewards/accuracies": 0.5103820562362671,
+      "eval_rewards/chosen": -0.0074609918519854546,
+      "eval_rewards/margins": 0.042396578937768936,
+      "eval_rewards/rejected": -0.049857571721076965,
+      "eval_runtime": 104.0533,
+      "eval_samples_per_second": 3.296,
       "eval_steps_per_second": 0.413,
       "step": 100
     },
     {
       "epoch": 0.29,
-      "grad_norm": 104.0,
+      "grad_norm": 68.5,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": 83.91990661621094,
-      "logits/rejected": 83.79493713378906,
-      "logps/chosen": -32.23216247558594,
-      "logps/rejected": -32.69822692871094,
-      "loss": 0.7793,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.6325147151947021,
-      "rewards/margins": 0.6663830280303955,
-      "rewards/rejected": -0.03386829048395157,
+      "logits/chosen": 83.77371978759766,
+      "logits/rejected": 83.6593017578125,
+      "logps/chosen": -32.28753662109375,
+      "logps/rejected": -32.78463363647461,
+      "loss": 0.7046,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.3676328957080841,
+      "rewards/margins": 0.43200382590293884,
+      "rewards/rejected": -0.06437097489833832,
       "step": 110
     },
     {
       "epoch": 0.31,
-      "grad_norm": 108.5,
+      "grad_norm": 67.5,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": 84.13631439208984,
-      "logits/rejected": 84.24298858642578,
-      "logps/chosen": -28.422576904296875,
-      "logps/rejected": -35.379127502441406,
-      "loss": 0.8738,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.40560561418533325,
-      "rewards/margins": 0.34159213304519653,
-      "rewards/rejected": 0.06401350349187851,
+      "logits/chosen": 83.89173889160156,
+      "logits/rejected": 84.00154113769531,
+      "logps/chosen": -28.253698348999023,
+      "logps/rejected": -35.52687072753906,
+      "loss": 0.7808,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.33794277906417847,
+      "rewards/margins": 0.37180477380752563,
+      "rewards/rejected": -0.03386203199625015,
       "step": 120
     },
     {
       "epoch": 0.34,
-      "grad_norm": 55.5,
+      "grad_norm": 47.5,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": 81.52787017822266,
-      "logits/rejected": 81.55335998535156,
-      "logps/chosen": -30.235469818115234,
-      "logps/rejected": -31.976581573486328,
-      "loss": 0.8013,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.5375091433525085,
-      "rewards/margins": 0.5734738707542419,
-      "rewards/rejected": -0.03596482425928116,
+      "logits/chosen": 81.15254974365234,
+      "logits/rejected": 81.17030334472656,
+      "logps/chosen": -30.407424926757812,
+      "logps/rejected": -32.09453201293945,
+      "loss": 0.8069,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.24996523559093475,
+      "rewards/margins": 0.3314167857170105,
+      "rewards/rejected": -0.08145156502723694,
       "step": 130
     },
     {
       "epoch": 0.36,
-      "grad_norm": 57.75,
+      "grad_norm": 44.5,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": 82.77738189697266,
-      "logits/rejected": 82.78349304199219,
-      "logps/chosen": -27.04022216796875,
-      "logps/rejected": -32.86003875732422,
-      "loss": 0.7387,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.32568174600601196,
-      "rewards/margins": 0.6750079393386841,
-      "rewards/rejected": -0.34932616353034973,
+      "logits/chosen": 82.38260650634766,
+      "logits/rejected": 82.38970184326172,
+      "logps/chosen": -27.169906616210938,
+      "logps/rejected": -33.05832290649414,
+      "loss": 0.7356,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.13870806992053986,
+      "rewards/margins": 0.4561801850795746,
+      "rewards/rejected": -0.3174721598625183,
       "step": 140
     },
     {
       "epoch": 0.39,
-      "grad_norm": 63.0,
+      "grad_norm": 50.5,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": 81.38068389892578,
-      "logits/rejected": 81.35902404785156,
-      "logps/chosen": -28.90557289123535,
-      "logps/rejected": -32.89466094970703,
-      "loss": 0.6488,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.38682785630226135,
-      "rewards/margins": 0.6493567228317261,
-      "rewards/rejected": -0.26252883672714233,
+      "logits/chosen": 80.86981201171875,
+      "logits/rejected": 80.84159851074219,
+      "logps/chosen": -28.839962005615234,
+      "logps/rejected": -33.071983337402344,
+      "loss": 0.628,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.2745725214481354,
+      "rewards/margins": 0.5273140668869019,
+      "rewards/rejected": -0.2527415156364441,
       "step": 150
     },
     {
       "epoch": 0.42,
-      "grad_norm": 66.0,
+      "grad_norm": 63.5,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": 83.2899169921875,
-      "logits/rejected": 83.30531311035156,
-      "logps/chosen": -33.436553955078125,
-      "logps/rejected": -30.43096351623535,
-      "loss": 0.6075,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.5601544976234436,
-      "rewards/margins": 0.9428972005844116,
-      "rewards/rejected": -0.382742702960968,
+      "logits/chosen": 82.75099182128906,
+      "logits/rejected": 82.76429748535156,
+      "logps/chosen": -33.66413116455078,
+      "logps/rejected": -30.465320587158203,
+      "loss": 0.7241,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.2363094538450241,
+      "rewards/margins": 0.4926998019218445,
+      "rewards/rejected": -0.2563903033733368,
       "step": 160
     },
     {
       "epoch": 0.44,
-      "grad_norm": 70.5,
+      "grad_norm": 59.0,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": 84.12855529785156,
-      "logits/rejected": 84.0693359375,
-      "logps/chosen": -30.63346290588379,
-      "logps/rejected": -32.29146194458008,
-      "loss": 0.7575,
+      "logits/chosen": 83.50555419921875,
+      "logits/rejected": 83.4540786743164,
+      "logps/chosen": -30.95868492126465,
+      "logps/rejected": -32.49937057495117,
+      "loss": 0.758,
       "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.5006591081619263,
-      "rewards/margins": 0.717779278755188,
-      "rewards/rejected": -0.21712002158164978,
+      "rewards/chosen": 0.15030263364315033,
+      "rewards/margins": 0.38995999097824097,
+      "rewards/rejected": -0.23965740203857422,
       "step": 170
     },
     {
       "epoch": 0.47,
-      "grad_norm": 65.0,
+      "grad_norm": 35.25,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": 81.81861114501953,
-      "logits/rejected": 81.79876708984375,
-      "logps/chosen": -30.488971710205078,
-      "logps/rejected": -31.519611358642578,
-      "loss": 0.6265,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.5665422677993774,
-      "rewards/margins": 0.8564668893814087,
-      "rewards/rejected": -0.289924681186676,
+      "logits/chosen": 81.02674865722656,
+      "logits/rejected": 81.00919342041016,
+      "logps/chosen": -30.531005859375,
+      "logps/rejected": -31.705917358398438,
+      "loss": 0.6438,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.3330736756324768,
+      "rewards/margins": 0.6074270606040955,
+      "rewards/rejected": -0.27435341477394104,
       "step": 180
     },
     {
       "epoch": 0.49,
-      "grad_norm": 34.0,
+      "grad_norm": 37.0,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": 83.52326965332031,
-      "logits/rejected": 83.50785064697266,
-      "logps/chosen": -30.285634994506836,
-      "logps/rejected": -30.630422592163086,
-      "loss": 0.8287,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.38462626934051514,
-      "rewards/margins": 0.5145466327667236,
-      "rewards/rejected": -0.12992039322853088,
+      "logits/chosen": 82.6941909790039,
+      "logits/rejected": 82.67329406738281,
+      "logps/chosen": -30.421396255493164,
+      "logps/rejected": -30.745235443115234,
+      "loss": 0.8436,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.172510027885437,
+      "rewards/margins": 0.3111144006252289,
+      "rewards/rejected": -0.13860435783863068,
       "step": 190
     },
     {
       "epoch": 0.52,
-      "grad_norm": 50.25,
+      "grad_norm": 35.0,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": 79.0837173461914,
-      "logits/rejected": 79.03185272216797,
-      "logps/chosen": -33.81468963623047,
-      "logps/rejected": -32.4266357421875,
-      "loss": 0.8098,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.7504364252090454,
-      "rewards/margins": 0.6970199346542358,
-      "rewards/rejected": 0.053416453301906586,
+      "logits/chosen": 78.1072006225586,
+      "logits/rejected": 78.06266021728516,
+      "logps/chosen": -33.60405731201172,
+      "logps/rejected": -32.724769592285156,
+      "loss": 0.6396,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.5743416547775269,
+      "rewards/margins": 0.690025269985199,
+      "rewards/rejected": -0.11568355560302734,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": 98.86815643310547,
-      "eval_logits/rejected": 98.84379577636719,
-      "eval_logps/chosen": -32.561100006103516,
-      "eval_logps/rejected": -36.16287612915039,
-      "eval_loss": 1.0544954538345337,
-      "eval_rewards/accuracies": 0.5278239250183105,
-      "eval_rewards/chosen": -0.09434036910533905,
-      "eval_rewards/margins": 0.06273789703845978,
-      "eval_rewards/rejected": -0.15707828104496002,
-      "eval_runtime": 103.8402,
-      "eval_samples_per_second": 3.303,
+      "eval_logits/chosen": 98.68441009521484,
+      "eval_logits/rejected": 98.66287994384766,
+      "eval_logps/chosen": -32.568355560302734,
+      "eval_logps/rejected": -36.311798095703125,
+      "eval_loss": 0.9231801629066467,
+      "eval_rewards/accuracies": 0.5834717750549316,
+      "eval_rewards/chosen": -0.06259084492921829,
+      "eval_rewards/margins": 0.11004550755023956,
+      "eval_rewards/rejected": -0.17263635993003845,
+      "eval_runtime": 103.7953,
+      "eval_samples_per_second": 3.305,
       "eval_steps_per_second": 0.414,
       "step": 200
     },
     {
       "epoch": 0.55,
-      "grad_norm": 104.0,
+      "grad_norm": 71.5,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": 81.58848571777344,
-      "logits/rejected": 81.49668884277344,
-      "logps/chosen": -33.169471740722656,
-      "logps/rejected": -34.991939544677734,
-      "loss": 0.7585,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.620996356010437,
-      "rewards/margins": 0.6256735324859619,
-      "rewards/rejected": -0.004677181597799063,
+      "logits/chosen": 80.72740173339844,
+      "logits/rejected": 80.64077758789062,
+      "logps/chosen": -33.17305374145508,
+      "logps/rejected": -35.227195739746094,
+      "loss": 0.6879,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.3863331079483032,
+      "rewards/margins": 0.5068855881690979,
+      "rewards/rejected": -0.12055248022079468,
       "step": 210
     },
     {
       "epoch": 0.57,
-      "grad_norm": 35.0,
+      "grad_norm": 53.5,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": 83.66175079345703,
-      "logits/rejected": 83.74690246582031,
-      "logps/chosen": -31.094005584716797,
-      "logps/rejected": -31.061717987060547,
-      "loss": 0.626,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": 0.683959424495697,
-      "rewards/margins": 0.9501722455024719,
-      "rewards/rejected": -0.2662127912044525,
+      "logits/chosen": 82.8826675415039,
+      "logits/rejected": 82.97186279296875,
+      "logps/chosen": -31.021907806396484,
+      "logps/rejected": -31.140172958374023,
+      "loss": 0.6275,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.46352285146713257,
+      "rewards/margins": 0.6691335439682007,
+      "rewards/rejected": -0.20561063289642334,
       "step": 220
     },
     {
       "epoch": 0.6,
-      "grad_norm": 69.5,
+      "grad_norm": 52.5,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": 80.7747802734375,
-      "logits/rejected": 80.82144165039062,
-      "logps/chosen": -32.322105407714844,
-      "logps/rejected": -34.30824661254883,
-      "loss": 0.7928,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.38691413402557373,
-      "rewards/margins": 0.6237919926643372,
-      "rewards/rejected": -0.23687779903411865,
+      "logits/chosen": 80.10808563232422,
+      "logits/rejected": 80.16153717041016,
+      "logps/chosen": -32.17018508911133,
+      "logps/rejected": -34.35881423950195,
+      "loss": 0.7509,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.31778058409690857,
+      "rewards/margins": 0.49111372232437134,
+      "rewards/rejected": -0.17333316802978516,
       "step": 230
     },
     {
       "epoch": 0.62,
-      "grad_norm": 89.5,
+      "grad_norm": 65.0,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": 83.18328094482422,
-      "logits/rejected": 83.4581069946289,
-      "logps/chosen": -30.6195125579834,
-      "logps/rejected": -31.637020111083984,
-      "loss": 0.5475,
+      "logits/chosen": 82.51493835449219,
+      "logits/rejected": 82.7935562133789,
+      "logps/chosen": -30.67529296875,
+      "logps/rejected": -31.787023544311523,
+      "loss": 0.5481,
       "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.854840099811554,
-      "rewards/margins": 0.9388192892074585,
-      "rewards/rejected": -0.08397923409938812,
+      "rewards/chosen": 0.5063848495483398,
+      "rewards/margins": 0.6338765621185303,
+      "rewards/rejected": -0.12749171257019043,
       "step": 240
     },
     {
       "epoch": 0.65,
-      "grad_norm": 84.5,
+      "grad_norm": 58.75,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": 81.9409408569336,
-      "logits/rejected": 82.00200653076172,
-      "logps/chosen": -26.908878326416016,
-      "logps/rejected": -30.06099510192871,
-      "loss": 0.8063,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.5496150255203247,
-      "rewards/margins": 0.6778196692466736,
-      "rewards/rejected": -0.12820473313331604,
+      "logits/chosen": 81.21392059326172,
+      "logits/rejected": 81.27059173583984,
+      "logps/chosen": -26.883764266967773,
+      "logps/rejected": -30.1818790435791,
+      "loss": 0.7593,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.3560645580291748,
+      "rewards/margins": 0.4966358244419098,
+      "rewards/rejected": -0.1405712366104126,
       "step": 250
     },
     {
       "epoch": 0.68,
-      "grad_norm": 60.25,
+      "grad_norm": 37.5,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": 79.24473571777344,
-      "logits/rejected": 79.37559509277344,
-      "logps/chosen": -30.283336639404297,
-      "logps/rejected": -36.4005241394043,
-      "loss": 0.4689,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": 1.0221258401870728,
-      "rewards/margins": 1.2612489461898804,
-      "rewards/rejected": -0.23912319540977478,
+      "logits/chosen": 78.52751159667969,
+      "logits/rejected": 78.65741729736328,
+      "logps/chosen": -30.363637924194336,
+      "logps/rejected": -36.511695861816406,
+      "loss": 0.5428,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": 0.5986787676811218,
+      "rewards/margins": 0.8037135004997253,
+      "rewards/rejected": -0.20503473281860352,
       "step": 260
     },
     {
       "epoch": 0.7,
-      "grad_norm": 31.875,
+      "grad_norm": 35.0,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": 78.6611328125,
-      "logits/rejected": 78.69209289550781,
-      "logps/chosen": -30.992252349853516,
-      "logps/rejected": -31.770212173461914,
-      "loss": 0.6471,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.6293355822563171,
-      "rewards/margins": 0.8411690592765808,
-      "rewards/rejected": -0.21183356642723083,
+      "logits/chosen": 77.82562255859375,
+      "logits/rejected": 77.8570785522461,
+      "logps/chosen": -30.939916610717773,
+      "logps/rejected": -31.843109130859375,
+      "loss": 0.5855,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 0.4195052981376648,
+      "rewards/margins": 0.5883487462997437,
+      "rewards/rejected": -0.1688433587551117,
       "step": 270
     },
     {
       "epoch": 0.73,
-      "grad_norm": 100.0,
+      "grad_norm": 64.5,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": 81.2646255493164,
-      "logits/rejected": 81.05351257324219,
-      "logps/chosen": -31.10286521911621,
-      "logps/rejected": -29.866031646728516,
-      "loss": 0.7333,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.5656233429908752,
-      "rewards/margins": 0.7248517274856567,
-      "rewards/rejected": -0.15922844409942627,
+      "logits/chosen": 80.51162719726562,
+      "logits/rejected": 80.30082702636719,
+      "logps/chosen": -30.975082397460938,
+      "logps/rejected": -29.75954818725586,
+      "loss": 0.7416,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.417405366897583,
+      "rewards/margins": 0.4636818468570709,
+      "rewards/rejected": -0.04627644270658493,
       "step": 280
     },
     {
       "epoch": 0.75,
-      "grad_norm": 61.25,
+      "grad_norm": 42.5,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": 81.45980072021484,
-      "logits/rejected": 81.37715148925781,
-      "logps/chosen": -33.03575134277344,
-      "logps/rejected": -32.42512512207031,
-      "loss": 0.6231,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.8547972440719604,
-      "rewards/margins": 1.1814041137695312,
-      "rewards/rejected": -0.32660672068595886,
+      "logits/chosen": 80.64032745361328,
+      "logits/rejected": 80.54534149169922,
+      "logps/chosen": -33.00518035888672,
+      "logps/rejected": -32.465091705322266,
+      "loss": 0.5643,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": 0.5495337247848511,
+      "rewards/margins": 0.7736457586288452,
+      "rewards/rejected": -0.22411206364631653,
       "step": 290
     },
     {
       "epoch": 0.78,
-      "grad_norm": 78.0,
+      "grad_norm": 46.75,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": 77.12594604492188,
-      "logits/rejected": 77.21607971191406,
-      "logps/chosen": -32.209590911865234,
-      "logps/rejected": -29.23952865600586,
-      "loss": 0.6965,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.9568928480148315,
-      "rewards/margins": 1.066897988319397,
-      "rewards/rejected": -0.11000506579875946,
+      "logits/chosen": 76.24769592285156,
+      "logits/rejected": 76.3338851928711,
+      "logps/chosen": -32.11992645263672,
+      "logps/rejected": -29.312702178955078,
+      "loss": 0.6566,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.642888605594635,
+      "rewards/margins": 0.748228907585144,
+      "rewards/rejected": -0.1053403839468956,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": 98.90528106689453,
-      "eval_logits/rejected": 98.8800277709961,
-      "eval_logps/chosen": -32.50230407714844,
-      "eval_logps/rejected": -36.192344665527344,
-      "eval_loss": 0.9868877530097961,
-      "eval_rewards/accuracies": 0.5830564498901367,
-      "eval_rewards/chosen": -0.047303199768066406,
-      "eval_rewards/margins": 0.1333501785993576,
-      "eval_rewards/rejected": -0.180653378367424,
-      "eval_runtime": 103.7665,
-      "eval_samples_per_second": 3.305,
-      "eval_steps_per_second": 0.414,
+      "eval_logits/chosen": 98.71778106689453,
+      "eval_logits/rejected": 98.69977569580078,
+      "eval_logps/chosen": -32.561683654785156,
+      "eval_logps/rejected": -36.26484298706055,
+      "eval_loss": 0.9383498430252075,
+      "eval_rewards/accuracies": 0.5456810593605042,
+      "eval_rewards/chosen": -0.05925469473004341,
+      "eval_rewards/margins": 0.08990433067083359,
+      "eval_rewards/rejected": -0.1491590440273285,
+      "eval_runtime": 103.9955,
+      "eval_samples_per_second": 3.298,
+      "eval_steps_per_second": 0.413,
       "step": 300
     },
     {
       "epoch": 0.81,
-      "grad_norm": 59.25,
+      "grad_norm": 44.25,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": 84.09150695800781,
-      "logits/rejected": 84.11959075927734,
-      "logps/chosen": -30.127155303955078,
-      "logps/rejected": -32.29438781738281,
-      "loss": 0.7446,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.5891044735908508,
-      "rewards/margins": 0.6648039817810059,
-      "rewards/rejected": -0.07569954544305801,
+      "logits/chosen": 83.27408599853516,
+      "logits/rejected": 83.30269622802734,
+      "logps/chosen": -30.081134796142578,
+      "logps/rejected": -32.3824348449707,
+      "loss": 0.6944,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.39120084047317505,
+      "rewards/margins": 0.48253631591796875,
+      "rewards/rejected": -0.0913354903459549,
       "step": 310
     },
     {
       "epoch": 0.83,
-      "grad_norm": 60.25,
+      "grad_norm": 30.375,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": 81.67195892333984,
-      "logits/rejected": 81.67171478271484,
-      "logps/chosen": -30.566574096679688,
-      "logps/rejected": -29.08144187927246,
-      "loss": 0.5518,
+      "logits/chosen": 80.83680725097656,
+      "logits/rejected": 80.83592224121094,
+      "logps/chosen": -30.626052856445312,
+      "logps/rejected": -29.309789657592773,
+      "loss": 0.5574,
       "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.8255898356437683,
-      "rewards/margins": 0.9551518559455872,
-      "rewards/rejected": -0.12956194579601288,
+      "rewards/chosen": 0.48625460267066956,
+      "rewards/margins": 0.6814028024673462,
+      "rewards/rejected": -0.19514815509319305,
       "step": 320
     },
     {
       "epoch": 0.86,
-      "grad_norm": 58.0,
+      "grad_norm": 37.0,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": 78.98292541503906,
-      "logits/rejected": 79.03089904785156,
-      "logps/chosen": -29.0233211517334,
-      "logps/rejected": -32.939231872558594,
-      "loss": 0.5283,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 1.063429832458496,
-      "rewards/margins": 1.306926965713501,
-      "rewards/rejected": -0.24349698424339294,
+      "logits/chosen": 78.04208374023438,
+      "logits/rejected": 78.09200286865234,
+      "logps/chosen": -29.09530258178711,
+      "logps/rejected": -32.94050979614258,
+      "loss": 0.5255,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": 0.6286521553993225,
+      "rewards/margins": 0.7814761400222778,
+      "rewards/rejected": -0.15282386541366577,
       "step": 330
     },
     {
       "epoch": 0.88,
-      "grad_norm": 88.5,
+      "grad_norm": 54.75,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": 83.09165954589844,
-      "logits/rejected": 83.11895751953125,
-      "logps/chosen": -32.228084564208984,
-      "logps/rejected": -33.656490325927734,
-      "loss": 0.6498,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.7872547507286072,
-      "rewards/margins": 1.0174915790557861,
-      "rewards/rejected": -0.23023685812950134,
+      "logits/chosen": 82.37565612792969,
+      "logits/rejected": 82.40386962890625,
+      "logps/chosen": -32.312721252441406,
+      "logps/rejected": -33.788272857666016,
+      "loss": 0.7038,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.4497138559818268,
+      "rewards/margins": 0.6595038175582886,
+      "rewards/rejected": -0.2097899168729782,
       "step": 340
     },
     {
       "epoch": 0.91,
-      "grad_norm": 42.5,
+      "grad_norm": 30.0,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": 82.23709869384766,
-      "logits/rejected": 82.24420928955078,
-      "logps/chosen": -32.544349670410156,
-      "logps/rejected": -33.209228515625,
-      "loss": 0.6454,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.8795096278190613,
-      "rewards/margins": 0.9734523892402649,
-      "rewards/rejected": -0.09394274652004242,
+      "logits/chosen": 81.36256408691406,
+      "logits/rejected": 81.37139892578125,
+      "logps/chosen": -32.320640563964844,
+      "logps/rejected": -33.50237274169922,
+      "loss": 0.5293,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.6615498065948486,
+      "rewards/margins": 0.866836428642273,
+      "rewards/rejected": -0.20528674125671387,
       "step": 350
     },
     {
       "epoch": 0.94,
-      "grad_norm": 66.0,
+      "grad_norm": 44.25,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": 83.5949935913086,
-      "logits/rejected": 83.62043762207031,
-      "logps/chosen": -28.472000122070312,
-      "logps/rejected": -31.736831665039062,
-      "loss": 0.6336,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.8641688227653503,
-      "rewards/margins": 0.9337539672851562,
-      "rewards/rejected": -0.0695851519703865,
+      "logits/chosen": 82.83390045166016,
+      "logits/rejected": 82.8671875,
+      "logps/chosen": -28.39254379272461,
+      "logps/rejected": -31.90033531188965,
+      "loss": 0.6161,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.579832911491394,
+      "rewards/margins": 0.7050749659538269,
+      "rewards/rejected": -0.12524208426475525,
       "step": 360
     },
     {
       "epoch": 0.96,
-      "grad_norm": 78.5,
+      "grad_norm": 70.0,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": 83.08488464355469,
-      "logits/rejected": 83.10858917236328,
-      "logps/chosen": -31.81646156311035,
-      "logps/rejected": -35.26220703125,
-      "loss": 0.6364,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.847761332988739,
-      "rewards/margins": 0.9700264930725098,
-      "rewards/rejected": -0.1222650408744812,
+      "logits/chosen": 82.28783416748047,
+      "logits/rejected": 82.31135559082031,
+      "logps/chosen": -31.76995849609375,
+      "logps/rejected": -35.398277282714844,
+      "loss": 0.6445,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.553102970123291,
+      "rewards/margins": 0.6975550651550293,
+      "rewards/rejected": -0.14445209503173828,
       "step": 370
     },
     {
       "epoch": 0.99,
-      "grad_norm": 82.5,
+      "grad_norm": 56.5,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": 77.07919311523438,
-      "logits/rejected": 76.94573974609375,
-      "logps/chosen": -29.726633071899414,
-      "logps/rejected": -28.218725204467773,
-      "loss": 0.7075,
+      "logits/chosen": 76.20292663574219,
+      "logits/rejected": 76.08332824707031,
+      "logps/chosen": -29.707965850830078,
+      "logps/rejected": -28.27734375,
+      "loss": 0.7106,
       "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.6795950531959534,
-      "rewards/margins": 0.7261168956756592,
-      "rewards/rejected": -0.046521805226802826,
+      "rewards/chosen": 0.43407946825027466,
+      "rewards/margins": 0.492464154958725,
+      "rewards/rejected": -0.05838475376367569,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.7616742146479619,
-      "train_runtime": 2551.1965,
+      "train_loss": 0.735349414874981,
+      "train_runtime": 2550.735,
       "train_samples_per_second": 1.207,
       "train_steps_per_second": 0.151
     }